# File lib/phusion_passenger/utils/unseekable_socket.rb, line 48 def wrap(socket) # Some people report that sometimes their Ruby (MRI/REE) # processes get stuck with 100% CPU usage. Upon further # inspection with strace, it turns out that these Ruby # processes are continuously calling lseek() on a socket, # which of course returns ESPIPE as error. gdb reveals # lseek() is called by fwrite(), which in turn is called # by rb_fwrite(). The affected socket is the # AbstractRequestHandler client socket. # # I inspected the MRI source code and didn't find # anything that would explain this behavior. This makes # me think that it's a glibc bug, but that's very # unlikely. # # The rb_fwrite() implementation takes an entirely # different code path if I set 'sync' to true: it will # skip fwrite() and use write() instead. So here we set # 'sync' to true in the hope that this will work around # the problem. socket.sync = true # There's no need to set the encoding for Ruby 1.9 because # abstract_request_handler.rb is tagged with 'encoding: binary'. @socket = socket return self end